Automatic Sublingual Vein Feature Extraction System

Abstract

The quintessence of the diagnosis in traditional Chinese medicine is syndrome differentiation and treatment. Syndrome differentiation consists of four methods: observing, hearing as well as smelling, asking, and touching. The examination of the observing is the most important procedure in the method of "tongue." In recent years, numerous medical studies have identified the close relations between sublingual veins and human organs. Therefore, sublingual pathological symptoms, as well as demographical information of patients, imply pathological changes in the organs, and early diagnosis is beneficial for early treatment. However, the diagnosis of sublingual pathological symptoms is usually influenced by the doctor's subjective interpretation, experience, and environmental factors. The results can easily be limited by subjective factors such as knowledge, experience, mentality, diagnostic techniques, color perception and interpretation. Different doctors may make different judgments on the same tongue, presenting less than desirable repeatability. Therefore, assisting doctors' diagnoses with scientific methods and standardizing the differentiating process to obtain reliable diagnoses and enhance the clinical applicability of Chinese medicine is an important issue. In its wake, this study aims to construct an Automatic Sublingual Vein Feature Extraction System based on image processing technologies to allow objective and quantified computer readings. The extraction of sublingual vein features mainly captures the back of the tongue and extract the sublingual vein area for feature expression analysis. Firstly, the patient's back of the tongue is photographed and color-graded to compensate for color distortion, and then the tongue-back area is extracted. This study extracts tongue-back imagery by analyzing the RGB color expression of the back of the tongue, lips, teeth and skin, translating it into the HSI color space easily perceived by the human eye, along with skin area removal, rectangle detection, teeth area removal, black area removal and control point detection. The captured tongue-back image goes through histogram equalization and hue shift to enhance color contrast. Sublingual veins are extracted through analyzing RGB color component shift, hues, saturation and brightness. Then the sublingual vein color information and positioning are used to differentiate hues, lengths and branches. Thinning analysis is used to determine the presence of varicose veins. At the same time, the surrounding features of sublingual veins, such as columnar vein, bubbly vein, petechiae and bloodshot, are extracted. The information regarding features and lingual vein conditions are integrated and analyzed for doctors' clinical reference. This study utilizes 199 lingual images for statistic testing and three lingual diagnostic experts in Chinese medicine for lingual reading. The accuracy for the extractions are: tongue back 86%, sublingual vein 80%, varicose veins 90%, branches 87%, and the accuracy rates for columnar veins and bubbly veins are 87%, 88% and 73% respectively.